ECOFEMINISMO DE AFINIDAD: ESPIRITUALISTA O

2.3.1 Method

Participants

26 (17 female) native British English speakers from the same population as Experiment 1 took part in this experiment. Their mean age was 22 years (range: 18- 34).6

6

65

Stimuli, Design and Procedure

All was the same as Experiment 1 except that the probe was presented at 300ms post sentence offset.

2.3.2 Results

Data preparation and analysis

Data were trimmed in the same way as Experiment 1 leading to three participants’ data and three sets of items being removed from all subsequent statistical analyses. The excluded item sets had low accuracy in the ambiguous-related probe condition; these probes (ambiguous word) were “layer” (coat), “metal” (cymbal), and “fireplace” (poker). Individual responses faster than 300 ms or made after the timeout (2500 ms) were also excluded; they contributed to 0.8% of the total data. The same analyses were conducted as in Experiment 1.

Accuracy

Like Experiment 1, overall accuracy was above 90%, for all probe types with the lowest accuracy for ambiguous-related and ambiguous-inappropriate probes (see Figure 2-2A middle column).

Sentence Ambiguity and Probe Relatedness. This ANOVA showed a significant

main effect of probe relatedness, where related probes were less accurate than unrelated probes (F1(1,18) = 29.5, p < .001, η2p= .621; F2(1,81) = 49.9, p < .001, η2

66

ambiguous sentence probes were less accurate than unambiguous probes (F1(1,18) = 4.54 p = .047, η2

p= .201; F2(1,81) = 4.82, p = .031, η2p= .056). The interaction between sentence ambiguity and probe relatedness did not reach significance (F1(1,18) = 2.66, p = .120, η2p=.129; F2(1,81) = 2.49, p = .118, η2p= .030), although, on average, the ambiguity cost was larger for related probes than unrelated probes (3% vs. 0.6% respectively).

Ambiguous Inappropriate Probes. The ANOVA comparing accuracy of the

ambiguous inappropriate probes with that of the ambiguous unrelated probes showed a main effect of unrelated probe type, where accuracy was lower for inappropriate than unrelated probes (F1(1,18) = 34.4, p < .001, η2p= .657; F2(1,81) = 28.3, p < .001, η2

p= .259).

Reaction Times

The correct RT data were transformed and analysed using the same method as Experiment 1. Again, for ease of interpretation, the means of the untransformed data are presented (see Figure 2-2B middle column).

Sentence Ambiguity and Probe Relatedness. This ANOVA showed no significant

main effect of sentence ambiguity, (F1(1,18) = 2.34, p = .114, η2p= .115; F2(1,81) = 2.12, p = .150, η2

p= .025), although responses were on average, 23 ms slower for ambiguous than unambiguous probes. This difference was significant in the untransformed data (F1(1,18) = 5.34, p = .033, η2p= .229; F2(1,81) = 4.49, p = .037, η2

p= .053). The main effect of relatedness was not significant in the transformed data (F1(1,18) = .062, p = .807, η2p= .003; F2(1,81) = .725, p = .397, η2p= .009) but

67

was significant in the by-items analysis of the untransformed data, where responses, on average, slower for related than unrelated probes (F1(1,18) = .363, p = .555, η2p= .020; F2(1,81) = 4.65, p = .034, η2p= .054). There was no significant interaction between sentence ambiguity and probe relatedness (F1(1,18) = .012, p = .914, η2p= .001; F2(1,81) = .214, p = .645, η2p= .003).

Ambiguous Inappropriate Probes. Comparing the ambiguous-inappropriate and

ambiguous-unrelated probe conditions, the results showed a significant main effect of unrelated probe type, where responses were 84 ms slower for the inappropriate probes (F1(1,18) = 13.4, p = .002, η2p= .427; F2(1,81) = 27.6, p < .001, η2p= .254).

2.3.3 Discussion

The pattern of accuracy across the different probe conditions was remarkably similar to that of Experiment 1 (see Figure 2-2A). Despite having an additional 200ms to process the sentence before the probe appeared, accuracy was still lowest for the ambiguous-related and the ambiguous-inappropriate probes. The ambiguous- inappropriate probes were, again, significantly less accurate than the ambiguous- unrelated probes, indicating a specific difficulty in suppressing contextually inappropriate meanings. Although the ambiguous-related probes also appeared to have particularly low accuracy, the interaction between sentence ambiguity and probe relatedness did not reach statistical significance. Despite this, there was still a significant main effect of ambiguity, suggesting that, even at this longer probe delay, participants still had more difficulty understanding ambiguous than unambiguous sentences. Together, these results show that these (relatively rare) comprehension failures are not eliminated by giving an additional 200ms before presenting the

68

probe. This pattern may suggest that disambiguation failures may be an evitable consequence of semantic ambiguities. However, before drawing such a conclusion, it is important to investigate whether comprehension improves with an even longer probe-delay that gives participants even more uninterrupted processing time to disambiguate the ambiguous sentence successfully. Experiment 3 examines this hypothesis with a 1 second probe-delay.

The pattern of response times, which examine trials where comprehension was successful, however, was markedly different to that seen in Experiment 1. Although responses were, on average, slower for ambiguous than unambiguous sentence probes, this main ambiguity effect was not statistically reliable and was only around a third of the size found in Experiment 1’s results (23 ms vs. 62 ms). Importantly, the interaction between ambiguity and probe-delay is significant, reported formally in Section 2.5. This suggests that, when probe words are presented at 300ms, the contextually appropriate meanings of ambiguous sentences are not significantly less integrated than those of unambiguous sentences. In contrast, the ambiguous- inappropriate probe condition still showed a specific, large (84 ms), and highly significant impairment compared to ambiguous-unrelated probes, suggesting that the inappropriate meanings are still more active than unrelated meanings. Taken together, these findings suggest a dissociation between the time course of integrating the appropriate meaning and that of suppressing the inappropriate meaning, such that inappropriate meanings are briefly maintained alongside the correct interpretation.

69

The finding that the 300ms probe delay produced well-integrated contextually appropriate meanings (or, at least, meanings that were not significantly less integrated than those of unambiguous sentences) suggests this process is relatively rapid and, hence, is efficiently enhanced by disambiguating information. The additional finding that this integration develops in spite of highly activated inappropriate meanings suggests that forming a new understanding of the sentence may not be substantially impaired by such inappropriate representations.

The finding that the inappropriate probes still incurred a significant cost with the 300ms delay indicates that meaning suppression during reinterpretation has two cognitive features. First, encountering contextual information that is inconsistent with a current interpretation is not sufficient for it to be quickly suppressed. This is directly consistent with the reordered access model of ambiguity resolution which claims that contextual information only acts to enhance the contextually appropriate meaning (Duffy et al., 2001; Duffy et al., 1988). It is also compatible with several other models of ambiguity resolution, including probabilistic constraint-based theories (MacDonald et al., 1994) and the structure-building framework (Gernsbacher, 1990; Gernsbacher & St John, 2001) that assume variable effects of context on the activation of an ambiguous word’s meanings. According to these models, contextual effects are dependent on its association with the meaning and the weighting of other information such as the meaning’s current level of activation. Second, the results demonstrate that activation of the appropriate meaning does not lead to simultaneous suppression of the inappropriate meaning. This is not predicted by models in which meanings are mutually exclusive such as Rodd et al.’s (2004) distributed connectionist model where increases in activation of the appropriate

70

meaning must necessarily correspond to immediate decreases in activation of inappropriate meanings. However, this finding may be accommodated if there are multiple levels of semantic representations such that word meanings can be activated at an early lexical-semantic level of representation that is separate from the subsequent sentential representation in which only one meaning is currently integrated. Multiple types of representations are explicitly incorporated in the structure-building framework (Gernsbacher, 1990) and have been formally and successfully implemented in computational simulations of ambiguity resolution (Gernsbacher & St John, 2001).

Furthermore, the finding that the inappropriate meaning was still active when the probe was presented at 300ms contrasts with previous research on the initial selection of an ambiguous word’s meaning. Such research suggests that meanings may be selected within 200ms of hearing an ambiguous word (e.g., Seidenberg et al., 1982; Swinney, 1979). Thus, the question remains, how long does it take to suppress the inappropriate meaning during semantic reinterpretation? Studies on initial meaning selection that used the same semantic relatedness task as this current study used longer probe delays and report that good-readers can suppress inappropriate meanings by 750 ms or 1 second after reading an ambiguous word (Gernsbacher & Faust, 1991; Gernsbacher et al., 1990). Based on these studies, Experiment 3 presented the probe 1 second after the disambiguating word.

71